Leveraging AI

282 | How to Turn Claude Into Your Automation Engineer and Ship n8n Workflows with Jeremy Grandillon

• Isar Meitis, Jeremy Grandillon • Season 1 • Episode 282

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 46:13

SECURE YOUR SPOT FOR THE 

MULTI-AGENT ORCHESTRATION AI COURSE: https://multiplai.ai/multi-agent-orchestration-course/

AI BUSINESS TRANSFORMATION COURSE: https://multiplai.ai/ai-course/ 

You're still building automations the hard way, or even worse, not building them at all, because it is too complex

Dragging nodes. Debugging JSON. Googling error messages at midnight. Meanwhile, the businesses moving fastest right now are handing Claude a plain-English description of what they need — and getting a production-ready n8n workflow back in minutes.

This isn't theory. This is what Jérémy Grandillon does every single day for revenue teams across Europe.

In this session, Jérémy will walk you through the exact process he uses to go from a business problem to a fully deployed n8n automation — using nothing but Claude. You'll see the actual conversation, the prompts, the back-and-forth, and the finished workflow. No slides. No fluff. Just a live build from start to finish.

The specific use case: a lead nurturing system. Someone downloads your lead magnet, and an automated sequence kicks in — scoring the lead, personalizing follow-ups, and triggering the right message at the right time. The kind of workflow that used to take a developer two weeks to build. Jérémy ships it in one Claude session.

Jérémy Grandillon is the founder of TC9, a GTM automation consultancy based in Paris. With 60,000+ LinkedIn followers and a reputation as one of Europe's leading voices on AI-powered revenue operations, he's a HubSpot for Startups GTM Mentor and the organizer of France's Clay Cup nomination events. When it comes to using AI to build the systems that drive pipeline, Jérémy is the real deal.

In this session, you'll discover:

- How to describe a business workflow to Claude and get a working n8n automation back — no coding required
- Why old-school automation tools like n8n are more relevant than ever in the age of AI agents
- The exact prompting approach Jérémy uses to get Claude to build, troubleshoot, and fix n8n workflows
- How to build a complete lead nurturing sequence — from lead magnet download to personalized follow-up — in a single session
- When to use rigid automation vs. AI agents (and why the answer is usually both)
- How to handle compliance, data privacy, and production-grade reliability when AI builds your workflows
- The tools and setup you need to start building automations with Claude today

If you've been doing automation the manual way — or worse, not automating at all — this episode will change how you think about what's possible.

About Leveraging AI

If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!

Isar Meitis

Hello, and welcome to another live episode of the Leveraging AI Podcast, the podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This is Isar Metis, your host and building. Workflow automations is an extremely powerful skill. It allows you to take manual work across multiple tools and tech stack and whatever, and automate them perfectly. Now, many people think that AI or AI agents are required in order to build such automations, but the reality that's not the case. I started using Zapier in 2015. That's 11 years ago. That's way before. We had access to AI as normal people. I mean, Google had it, but we didn't. And but, but now with the introduction of AI tools, these automation capabilities became a hundred x more powerful. And the reason they're so much more powerful is because now you can benefit from the rigid step by step nature of the automation and combine it with the AI's ability to analyze, create, process, et cetera, do all the things that AI knows how to do, and then you get the best of both worlds. You get the. Rigid, solid, consistent output of an automation, plus the smart capabilities of ai, plus the AI's ability to, if you want, direct different channels of the automation. Uh, based on what he has analyzed and the exact situation now, developing sophisticated workflow automations was until not too long ago, a highly sought after scale and people charged a lot of money to either build it for you or teach courses on how to do this and. Tools, you know, like NA 10 and Make, and Zapier and all these kind of tools. The reality is right now, Claude knows how to build extremely powerful NA 10 automations on its own and in order. To be able to do that. Like it's, and it's doing this in order to be able to complete tasks that it doesn't know how to do on its own, or I should say it doesn't know how to do on its own yet. So all the universe that NAN has of connectivities and automations and things like that still do not exist in the cloud world. And Cloud knows how to build these processes for itself so it can connect and do things that it just can't do out of the box. Now in today's episode, we're going to show you exactly. How you can use cloud to develop really complex automations without knowing anything about NA 10 or code or anything else. Our guest today, Jeremy Grandon, he is the CEO of TC nine, which is an agency that designs and builds AI automation powered go to market engines for B2B companies. He's been doing this for a living for about three years now, so he's built. Probably hundreds of these automations. And for a very long way he had to actually build it himself. And now he uses Claude to build it and he builds it for companies for live production solutions. So he is not just toying around, he's going to share with us exactly how to build an automation like this, and. To be more specific, he's going to show us how to build a lead nurturing process beginning to end using this methodology. So knowing how to build lead nurturing processes is important by itself. Knowing how to build any automation you want using cloud of NA 10 is obviously a lot more powerful because you can then automate anything in your business. And hence, I'm really, really excited and honored to have Jeremy at the show again for the second time. Jeremy, welcome back to Leveraging ai.

Jeremy G

Thank you very much. Thank you. I'm very excited. That's a, that's a great intro. Uh, that's exactly what we are going to, to break down today, so I'm very excited to share that with you guys.

Isar Meitis

Yeah, I, I'm, I'm really excited. You know, you and I exchange ideas on LinkedIn a lot, and I love following the stuff that you're doing, and we're in the same space of building really cool stuff, and it's, I think the world right now is really crazy. Like if you wanna build something, you can, it's really just,

Jeremy G

yeah. Yeah, exactly. And it's going, it's going very, very fast. We were talking about that, uh, a few days ago talking about cloud tropic release. Like I think everything, every, every day last month. A new feature every day. Yeah. So it's

Isar Meitis

going, they released 74 things in 52 days. It's like, nah.

Jeremy G

Yes. And not small things like obviously some are, are small, but most of them are quite, quite impactful features, so yeah, it's, it's going very, very fast. That's a, that's a exciting period of time right now.

Isar Meitis

Yeah. with that, uh, let's get started. Jeremy, it's, it's, it's your show. So show us the magic.

Jeremy G

Okay, let's go. Um, so before, before diving into the, the, the technical things, uh, maybe it's. Let's put the stage and, uh, and, and give some context. So as, as you mentioned, I'm working in, in AI automation for, for a few years now, um, for GTM purposes. So basically, basically the goal is to automate the processes that are generating, generating revenue, so mostly marketing, sales and rev ops processes. So that's all our expertise at TC nine. And with that. I have an example today to share with you. You mentioned the lead nurturing. So the scenario is you create a lead magnet, so that's more a marketing scenario. In that case, you create a lead magnet. So good piece of content that you want to, you know, gate, uh, put behind a gate to collect contacts. But in the marketing world, you the more question you ask. The more friction you add, the less chances you have to for people to, you know, give the contacts. If you're asking too many questions about them, they might, you know, just keep it and say, ah, that was a good resource. But next time, so we want to remove this friction as much as possible. So we only ask for an email. But the problem with that is that one. You have to find who is this person. And two, you have to, you know, qualify and, and reach those contacts and make sure that you're not using the personal email because of, you know, regulations like GPR, especially in Europe. If you want to use those contacts for sales pur purposes, you have to find a business email. So that's all those, you know, constraints that I just described that we're going to solve with this automation. To not only talk only to the right people, but make sure that we are doing this within the, the, the, the regulation and the rules. So with that said, I've prepared, um, three things to share with you. The first thing is I will break down the entire workflow and show how important the, actually the, the idea, the, the, the process in your mind is the value now because all the rest I'm going to share with you. Has been done with ai. So the very starting point is, okay, I have a clear idea. And of course you will iterate, iterate with code, code and, and you will find new ideas and improve it. But if, if your process in the beginning is not clear for you, you will probably struggle. So that's the very first thing. So I've created a, a visual, so you can see for those who are just listening, I would describe. Then I will show you the final output, which is the big NA 10 workflow that has been created. With me, not by me. And then I will show you how I did it in a, at least a piece of it, uh, how I did it in code. Code.

GMT20260406-163401_Recording_avo_1280x720

Sold is brought to you by training courses by Multiply, which is my company. We currently have registrations open for two different courses. The first one is the AI Business Transformation course. It is a course we have been teaching since April of 2023. So for three years now, at least once a month, thousands of business people have learned AI fundamentals through this course. So if you are looking to build a solid foundation and learn multiple business use cases as well as tools and exactly how to apply them, including data analysis, writing proposals, creating videos, images, and so on, basically solid fundamentals across the board for AI usage in the business world. This is the right course for you. I am the one teaching this course over Zoom. So it's actually me, not an AI avatar. And it's not a recorded session. It is me. You can ask questions, you can interact with other people like you. And the coming cohort starts on April 20th, and it goes for four weeks in a row. Every Monday, two hours every single time. And in the middle of the week you can come and join us for our AI hangouts and ask questions in between the sessions as well. It is probably the best course out there to give you solid AI fundamentals if you want to learn how to apply AI in your business. The other course is our Multi-agent orchestration course. This is a more advanced course that teaches how to build, like the name suggest multi-agent orchestration solutions using cloud, cowork and cloud code integrated with other tools integrated with your entire ecosystem and pick stack. And so if you are a little more advanced and you wanna start your way into the agent era and be able to automate literally any digital work in your business, build entire teams of employees, come and join that course. We have launched the early bird registration last week, and it is almost filled up to early bird sessions when I was planning to do just one. But the demand was absolutely crazy. So you might make it into the early bird session, but if not, we are opening another regular cohort just the following month. You can find out more information and register for any of these courses, either one or the other, or both if you want to, because you can start with the basic course in April and then join the other course later and you can find all of that information in, links in the show notes that will take you to the specific courses. And now back to the episode.

Isar Meitis

Fantastic. I'll say two things about what you said, because I think they're very, very important. One is the idea side of it and, and the other is how to develop an idea in a better way. So the idea side of it, I think, and I'm a really big believer, and I said that on the podcast several times in the past few weeks. I think the limiting factor of businesses used to be resources. How many people, how many, how much money in the bank, how many computes, how much whatever. And right now it's three things. It's having good ideas. Having good judgment and having good requirements, documents and definitions, uh, which connects exactly true to to what you said and what I've done, like the very first multi-agent thing that I built together with Claude is, is a tool that, that interviews me about the idea to really understand the essence and, and push me to. Instead of idea like, okay, if you wanna productize this, you need to know the target audience. You need to know which technology you're using. You need to know how frequently you're gonna use it. Are you gonna use it yourself? Are you gonna sell it to other people? Like it's gonna ask me,

Jeremy G

yeah,

Isar Meitis

about 30 to 45 minutes worth of questions that I need to answer. And then it writes, A PRD writes up 30 to 50 page product requirements documents that then Claude knows how to build, which is really, really awesome.

Jeremy G

Exactly.

Isar Meitis

And so

Jeremy G

no, exactly.

Isar Meitis

Knowing your idea, coming up with good ideas and knowing how to define the requirements for them are the new superpower.

Jeremy G

A hundred percent. A hundred percent. And you know, in our business, we, we are building those automation Now. Cloud code builds the automation. So we, we could think like, oh shit, that's the, that the end of our business. But actually we're very confident because code code won't replace what you described, like our expertise, our experience. I have. Almost nine years experience in sales and marketing. That's, that's the thing. And also adding to our offer some, some, you know, some consulting, some cons, consulting advice, um, where to start, how to do it right, or ev all the steps that you described. I think our business is going to, to be safe for, for a few more years. So, so, so that's, that's the very important point. Alright, cool. Let's, uh, let's dive into it. Let me share my screen

Isar Meitis

yep.

Jeremy G

Okay, cool. So this is the idea. So that's the workflow I describe. I will go through this quite, uh, quite fast, but the very first step and I will describe everything for those while just, uh, listening. As I mentioned, the first step, the intake is the visitor download the magnet. So they gave us an email and they have received the, the resource that's very classic for, for marketing, uh, workflows. And that's the trigger. So when we collect this, by the way, I created the, the gated tool. So we basically. Grab this from our website and created all those, those, uh, those steps through cloud. But that's for another, another episode. Then I re, I saved this, this, uh, this email in a, in a database and that triggers our workflow. So then the second step is the identity resolution. So as I mentioned, the very first thing we want to know is, is this a personal email or a professional email? We are not going to, um. To the same steps. If it's a personal email, we want to find a professional email attached to this personal email. And if it's a professional email, then we continue and go to the next step. Um, note on that also one value, because if we zoom out a little, we could imagine, we could say, okay, why I'm not, you know, automating everything with cloud code. Why am I using NA 10? But as you said, for maintenance. Not all the steps are requiring ai. So you can save in resources, you can be more consistent. So all those reasons are why we are using, uh, workflows, uh, and tools like NA 10. And in that

Isar Meitis

scenario, yeah, I'll say, I'll say two things just to add to this. I think the, the two places where, where the three. Kinda like decisions whether you wanna build on automation or just do it inside of cloud code. One is, is it a rigid step-by-step process? If it's a rigid step-by-step process, a NA 10 will do it cheaper. B na 10 will do it consistently. You give, you build an agent. An agent will do it. The same way 90% of the time. But that might not be enough. You may want it to run a hundred percent of the time, and NA 10 will run the process a hundred percent of the time, exactly the same. Uh, and then the last component is really the, the access to other tools, right? So the one thing that Claude still doesn't do great is. Built in connectivity to any tool you want. And there's MCP, but it's just not as efficient. And so having access to the entire universe that NA 10 has access to, uh, just gives cloud code through NA 10 access to all that stuff. So it's, it's a very useful from all these reasons.

Jeremy G

Yeah. And it, and it is also safer because if you are building everything into cloud, you are not necessarily a developer. And you might make mistakes in terms of security. Um. Relying on, on NA 10, they have those, those topics already. Yeah. So that's also safer. Um, so yeah, good reasons to, to continue using, uh, NA 10. And when you choose a, you think of a, a workflow and you choose a tool to support it. For me. Now, the MMCP connectivity, the API or how open the API is. Uh, is a new criteria. So N8N is very open, so you can connect it into a lot of things into NA 10. Some other tools won't be that open. So that's also a criteria for you, uh, in the future if you want to think about. Yeah. Okay. Is it agent ready? More or less? So, yeah, to continue with the, the, the, the workflow. So we call in that case. In NA 10 APIs to other tools. So that's goes with your point. Uh, you just mentioned in that case, food language, which is, which is a great, um, tool to find professional contacts from a personal, uh, contact. And based on what it returns, then we go in the direction on another, we have all the errors You. Um, pass. But if it goes well, then we find the person and then we start the end reaching. So in that step, we find the name, full name. So from just an email, we find the name, title, LinkedIn, uh, profile company. So that's all we need to continue and enrich the rest. And then we continue with our criteria. So that's step one. Is it personal or per or professional? Step two, who is this person? And once we have find, find them step three. Okay. Let's test them against our criteria. So the first one is the CRM check. Is this contact already in our CRM based on the, on the, on the email and the LinkedIn profile? If yes, just update me. I just want to to know, okay, this contact has been done downloading your, your resource. So maybe they are, you know, looking for. Extending, you know, uh, renewing, whatever. So that's a, that's a good signal you want to capture anyway. So we, uh, update our HubSpots, uh, in our case and trigger Slack notification so we can see it in a, in a channel. If it's a new contact, then even better, we continue and we push them to more filters, and that's the enrichment filters. So that's where we collect everything. We can, you know, contacts and everything. And we qualify the person. So that's where the AI enter the game in an eight 10. So we have a first qualification step, which is on the person. Is this person matching our ICP? Um, you know, is, uh, is this person in the right, uh, location, geographic location, for example, is this person, uh, ha does this person have enough, uh, experience? Uh, in it, in its role, et cetera, et cetera, those type of criteria, that's an AI agent at this stage. If no, we save everything we have done so far in our CRM, but we disqualify it. That's a person is interested in what we do, but it's not a potential client. Maybe in the future, that's why we save them, but not now. Then we continue and we qualify on the company level. So now that we have identified, okay, this person is potentially a decision maker, is the company interesting for us? Is it, uh, big enough? Is it in the right location? Is it, you know, in the right industry, et cetera? Your criteria, same logic. If you disqualify here, we save them. Maybe in the future, this person would change job and we monitor our contacts. So that's always good information to, to keep. We have many of those workflows working in the backend. But in that scenario here. If no, we just save it. If yes, we continue and we go to the next step, which is the routing. So that's basically updating the CRM. But most importantly, once it's updated with a new contact who is interested, who is qualified in our CRM, the next step for us is to add them into a sequence to send them a message, and of course, send the slack notification for the little dopamine it, uh, for us in the team. But then that means that from one person downloading a lead magnet on our website or whatever on socials, we automatically qualify them and reach them, save them in our CRM and start a conversation with them. And that's works behind the scenes

Isar Meitis

and Fantastic. I, two quick questions. One, what was the tool that you started with in the beginning to get their information, like the very first information you're getting about them?

Jeremy G

Yes. Uh, full. Enrich This.

Isar Meitis

Full enrich, that's the name of the

Jeremy G

tool. Yes. Yes. Full Enrich is

Isar Meitis

the, and it's a, it's an API And you pay per usage or something like that?

Jeremy G

Yeah, you can have a subscription and it will, you know, consume your subscription or you can, uh, use, uh, directly with, uh, per cost, I think. Yeah. A to confirm.

Isar Meitis

Cool.

Jeremy G

I use

Isar Meitis

this with. Very straightforward, right? It's exactly what a human would've done. Like, you check if that's a relevant person. Do we already have him in the CRM? Uh, if we don't have it in the CRM, okay, how, what other information can I find about that person then does that, now that I have the information, does, is it aligned with my ICP or not? If it is, awesome, if it's not okay, I'll save him maybe later and, and then you send them a message. The last question that I have, do you personalize the message in the sequence based on what you learned, or is the sequence the same sequence every time?

Jeremy G

Yeah, so it is personalized on a lower level. Uh, for now, because I'm preparing a new one, which is way more advanced than, it's taking a bit, a bit of time, but right now it's just, uh, I'm sending a voice note on LinkedIn to this person and just asking, you know, how did you find this resource? Was it helpful for you? Should we, should we talk about. Doing more, something like that. And

Isar Meitis

the voice note is, is the same voice note or is it like, uh, 11 Labs voice with you saying the name and stuff like that?

Jeremy G

No. So right now it's a, it's a generic one, so I'm just saying, Hey, how was the resource? You downloaded it? Uh, that's, that's the, that's the thing I have on my, on my to-do list, to use the elephant lab to, to do the, you know, dynamic one. That's, uh, that's exciting.

Isar Meitis

Okay. Awesome. So let's look at the final outcome. How does this thing work?

Jeremy G

Yes. So from those steps that are looking pretty, you know, simple. Let's say we go to something like this. So for those who are not seeing the, the screen, uh, that's, uh, N 10 Workflow. I will zoom in. Uh, but that's, um, uh, 59 notes. So. You can imagine, uh, how long it would've taken me to create and set up those 99. Notes manually

Isar Meitis

create and test. I think that's usually the

Jeremy G

And test that's even, yeah. Yeah. Even be a bigger point. I agree. So if I zoom in a little, that will reflect what we just talked about. Like this is the web book. I also have, you know, some manual for the testing for manual entry. Some, uh, bulk entries for the ones that I missed before I created this, uh, this workflow. Then normalization classified email, full language research. I would, I would go quite fast on, on those, on those things because it's, you know, not the sexiest those nodes, con conditions, you know, passing the results, et cetera. Slack notification, as you can see here, some timers like that. This, this thing was interesting when I, I was creating this with my agent because I was like, yeah, let's just do the, you know, check the, check the database in full enrich m in that case, and then move to the next step. Uh, but. I didn't count that it has to call the API. That takes time and then the result has to come back to the workflow. That also takes time and that, and it said, oh, you should add a, you know, wait step here for like 30 seconds, otherwise you won't receive anything. I was like, Hmm. Good point, good point. Thank you agent.

Isar Meitis

I, I, I'll say something about this, like when, when, back in the day, in December of 2025, when we, when we used to create these, when

Jeremy G

we were doing this,

Isar Meitis

we were, we would save, like there was a whole process on handling issues and mistakes in the process so you can document it and learn from it. And now it just does it, it's just, you don't even have to track anything. It just knows what to do and how to fix things and, and things like that, which is, which is really, really amazing.

Jeremy G

Exactly. Exactly. And I will show you the. One of the conversation I had with, with this agent, actually, the first one, which I, I could say, uh, that's, that's, yeah, that's impressive. I, I will show you that right after. But anyway, that's the, that's the, that's the, that's the workflow that, that are the agents that qualify the person to qualify the company. It also documented like little, you know, comments on the, on a single in case, I don't remember what's, what's what,

Isar Meitis

one, one pause on this. So all these tools, whether you're using MAKE or N 10 or, or, or, or. Uh, relevance or whatever. All of these tools have built in agents, and the trick is to know. When to use them and you want to use them when you need to, in this case, qualify or analyze information, right? Everything else runs rigid, step by step by step. And then the agents is but, but Claude also knows how to build the agents. And so you don't really need to know even what that means. But an agent is basically just a set of instructions connected to an AI brain behind the scenes where it can go and do something. And in some cases it's connected to tools to actually take some actions as well. But in this case, it's purely an analysis step.

Jeremy G

Yes. Yes, exactly. Good. Uh, good point here. Uh, yeah, we feed it with the information that comes from the previous step and say, okay, this is the prompt we need, you know, this talk criteria. Is this person doing this, this, and this? They use the information of the per on the person and the, and the, and the qualifying and yeah, uh, HubSpots steps, slack steps. So as you can see, it's 59 nodes. Pretty, pretty big stuff added. Did in a few, I would say. Days, I would say days like, uh, two, three days, something like that. And I will show you right now, which, which for, for a complete reason would've probably taken me a few weeks to do it properly with testing and, and, and all of that. So yeah, it's a, it's a great saving time. Uh, let me show you and,

Isar Meitis

and I'll say something about what I said before, knowing how NA 10 works helps a lot, but it's not mandatory anymore. The fact that Jeremy has all that experience of building any 10 processes before helps a lot because you understand the logic, you understand how it works when he gets stuck. I didn't get to a point where he got stuck and he didn't know how to solve a problem. I did get to a point that it got stuck and spend two hours on something that I can solve in six seconds because I know it's broken. And when, when it's close, when it's like, okay, it's gonna take it eight minutes and I can do it in six seconds, okay, I will later run the eight minutes because I'm doing something else. But it's when it's stopping me two hours from completing the process. I'm like, okay, this is ridiculous. I, I know what to fix. And so. Knowing how N 10 works and knowing how these tools work really helps, but it's really not necessary anymore because it will solve the problems that it run into sometimes in a the right way and sometimes in not the optimal way, but it will still solve the problem.

Jeremy G

Yeah, a hundred percent. And, uh, yeah, I think the, the, the cliche that says like, uh, you know, inputs, quality of the input equals quality of the output is still true. The, the, the, the ai LLM will be as good as the context you give it. And if you add your own expertise, you will be in control more of what's going on, and you will, you know. Save probably some tokens. That's another conversation. And, um, and some time, uh, a hundred percent. Yeah, that's actually a good transition before I show you the, the conversation that's, uh, how to build an agent. Uh, so very, very short and summarized version. You want to create files, MD files, and in those files have. Rules,

Isar Meitis

uh, pause. Pause you just for one second. MD files are markdown files, which is basically think about Microsoft Word without. The fancy design. Right, exactly. It's just pure text. Exactly. Uh, and this is how most of this universe of agents works, so it just, the most efficient way for it to read and write stuff is just writing, reading and writing markdown files. But it's just text. Like, it's nothing fancy. You can open it and read it yourself. It's written in English or in whatever other language you want, but most of them are in English. Uh, and, and that's it. It's, it's just a, it's just a set of instructions in English.

Jeremy G

Hundred percent. And, uh, that's actually great to add this. And the, the, the value of those D files is as, as you said, that there is no fluff and, and fancy visual things. So it's very light. So you can have your this, because basically an agent lives into, into a folder when you, where you put those files to guide it and say, okay, this is your identity, this is your mission, and this is your basically skills and tools. And that's all described in those files. And from there, every time you start a session, it'll go through those files in a specific order and be ready to, you know, use those tools, use those skills, and et cetera. So that's how you create a, an agent. And that comes with what you said. If you can prepare your agent in advance, let's say, okay, this agent I will use for NA 10 only, then a good practice would be, okay, let's give it, you know, a purpose. Uh, in my case, I gave it a name. You will see, give it some personality. Uh, you can, you can give it, uh, tools, obviously the MCP to connect to your, uh, n and workflow. So there is a setup. Step to do, uh, but also you can give it knowledge. And I would recommend when you create an expert agent to, you know, let's say, okay, you are an expert in NA 10. The very first thing you have to do is to read all the documentation that is online and save the best practices and save, you know, critical bugs that are blah, blah, blah, everything. And it will save it into an empty fight that will reopen every time you start a new station. But that's also a good, a good thing to, to keep in mind. So you can increase its capacities basically. So with that said, let me show you the first conversation I had with, with my agent, and I think the very few first few message are quite, quite cool. Okay. Can you see it? So for those who are not seeing, I will, I will obviously describe, yeah. Okay, cool. So I called it 80 because it's NA 10 and I'm not very inspired with names as you can see. The very first prompt that I started, so as you understood, I started this conversation include code for the, in the desktop app on my, on my, on my MyBook Pro and in the folder that was, um, holding those files I was describing. So. This agent lives in a file, in a folder, sorry, with the files inside of it, and that's where you start the conversation. So basically I said the very first line is, Hey, let's work together, because I know that it's the first thing it'll do. It's to read its files and understand what it is, what he has to do, the mission, et cetera. So basically it read the, uh, the files. Let me boot up properly and ready. So I, I've done this, the setup, the technical setup, uh, before, but the MCP was connected. What are we building? So it already knows. I just said, Hey, let's work together. It already knows what to do, what we are going to, you know, we are building in NA 10 for building workflows. So my first response was to just test, because it was the, basically almost the first run. So I wanted to just, you know, can you see my workspace? Tell me. My, my, um, workflows that are available like this. So as you can see, it was a bit messy, some testing that I had, but it was correct. Let's see if you can create one from scratch. Basically testing. So you will go through those steps when you create your agents in the first time. But the interesting thing is that, okay, that's coming here for, so for those who are seeing, I will show more. I sent a huge and long prompt. That is basically the v one of my workflow that I shared before in the, in the visuals. So that's, as you can see on the screen, I, I, I saw it's very long. So I use the, you know, the voice capture

Isar Meitis

and let's read a few small segments so people kind of get an idea of what, what's included in there.

Jeremy G

Uh, so I started by saying, okay, let's work on a real workflow, because I was testing stuff before, so that was the beginning of the sales stuff. Uh, that's, this is a workflow that receives and qualifies leads for our business. The overview is that we share leads, magnets on different channels. People give their email, and from this email, we want to know who is this person, what their, what their jobs are, if they're qualify to be a potential client. They are detailed information, blah is to breakdown. We receive the address. Email address thanks to WebBook. So I already knew you a little about that in those email address. Some would be professionals, some would be personal.

Isar Meitis

So the, the reason I wanted you to read some of this is because to connect it back to what we said before, two things that we said before. One is the flow chart that Jeremy showed us, and second is his real. Experience that he said before that comes into this, right? This is a simple description of hiring a new employee and telling them how to do the process. Like a detailed SOP. This is all it is. Like people think there's some magical thing in creating agents. It's not. It's literally in a detailed way. And again, you can see this is a long prompt and this is just version one, just explaining. What's the data that we have? What are we trying to do? What is the goal? What is the output we're looking for? What format needs to be like, all these things, but just explained in simple English, like there is no secret sauce. The secret sauce is knowing in as much detail what you want to achieve, where the data resides, and what the output needs to be.

Jeremy G

A hundred percent. And, uh, as you can see in the in the room photos who have the video, it's pretty close to what I described in the, in the first part of the, of this, of this, uh, of this, uh, podcast, of this webinar. And I'm just describing my criteria, like Okay, the qualification. Yeah. Uh, I want them to be, uh, located in the US Canada. That's, that's part of my targeting, you know, qualified company. I want them to be. Below, um, not below 15%, for example, employees, et cetera, et cetera. I want also to avoid my competitors. I don't want to, you know, send a, send a potential, uh, sales message to a competitor, et cetera. So that's very, you know, um, I was just, and in that scenario, I was talking to my laptop because I was using the voice capture. So it's very coming naturally, uh, in that case. And that's it. I sent it. This, it did a lot of. Steps that I don't really know what happened in that case. And from there, like, okay, I've made a plan. Cool. It will be a workflow with 50, up to 55 to 60 nodes. Pretty accurate. It was 59 in the end, and explaining all, all the steps. And then it's, it's where the, the, the real work starts. Uh, when you use agent is like basically corrected and guide it. So for example, it said, oh, okay, we go through, we would use ALO in full language, but I was, I wanted, used to use full language and ly. So I said, oh, okay. No. Um, replace Apollo. Uh, so that's probably somewhere here Yeah. Replaceable for this, this, and this. But Full Enrich doesn't return LinkedIn very well because it's read the API documentation on the on the Full Enrich website and it could that, oh, you can't do that with Full Enrich. I was like, okay, cool. Good to know. I don't want to read the entire description of the API on the full Enrich website, but now I know that there is this missing point here, so let's use another tool for that. And that's, that's basically how you,

Isar Meitis

yeah. And I, and I wanna add something about this as well. Like you, you will see that once you start building these, you run into this all the time. Like a, you don't know which tools are out there, B, you don't know, which are free, which are paid, and C you don't know the pros and cons of each one, but the quote unquote internet does know, like there's chat rooms and there's the actual API documentations and so on. And literally saying, this is what I'm trying to achieve. Go and research tools and come back with recommendations of what we should use. Uh, and I literally. Doing this and probably Jeremy as well, almost every single day, unless there's tools I already know that are working and so on. But, but if not, I literally just ask it and it comes back with a summary and a recommendation. And in many cases I do not agree with the recommendations, but I have the information now to make a decision. Yeah. And like, okay, let's go with a free tool for this and just compliment it with the paid tool to do that. Because it's, the tool is not a subscription, it's just per token. So I can do most of it on the free tool, and then I want you to do the other last 10% with the pay tool. So it's these kind of, uh. Decisions that you still can make. You don't have to. You can just let it go and do it, and it will do it, but it will give you better outcomes, cheaper, faster, more tailored to your needs if you are bringing your expertise and your knowledge and your decision making into the process.

Jeremy G

Yeah, absolutely. And uh, and also to add to this, sometimes you want to, you don't care. You are in testing and or exploring mode. You are like, okay, I want to create this thing. Just do it. I don't really care what tool you use, what you know, technology you're using. Just, just do it. Just do it and make it work. And it would come up with the best version possible with the context you give it. And once you have something that's working and that's how you know, um. Fast. It unlocks things be before compared to before. You can have a an MVPA, a viable version of your product or workflow very, very fast. Test your idea and then say, okay, that's working. Now I'm going to replace this bot by this paid, uh, use this paid tool here because I know the data is better. Whatever, you know, that's, that's, that's a great way to iterate on, on a, on a new idea. Yeah. Uh, from, from there, basically you will iterate. I'm probably, uh, to, uh, probably about, you know, version 10 of this workflow because I started and I iterated. So that was the very first idea. And then I was like, oh, damn, if you, maybe you have noticed in the first prompt, big prompt, I forgot the hubs spots. Control is this contact already existing? And I was, I realized, oh damn, it's not, it's not here. So let's add this and et cetera. So you iterate a few times and then you test testing is very, the big part of it. And that's how you create, create in a few hours what you could, uh, you, you would have done in, in a few days before. And not, not so long ago. I, I agree. Like end of 2025.

Isar Meitis

Yeah. Uh, something about the testing, which I think is important, and again, this is more thinking from like an engineer versus, uh, anything else. Knowing what to test and when to test and how to test it. And it's something you will develop if you've never done this, but I ran software companies most of my life, so I have, it's part of my DNA, I didn't actually do the testing. I had people doing the testing, but it doesn't matter. It's the, the idea of. Starting testing specific components and sometimes it gets stuck like it doesn't know. I'll give you a crazy example that has to do with Claude, not specifically with N 10, but it's a great example. Like I now build a lot of. Claude plugins, and we're not gonna dive into this, but plugins are, think about bigger packages. They have multiple scales and agents and connections and all these kind of things packaged into a package and, and I was in version whatever, 3.4 of the plugin and it wouldn't install. It got stuck again and again and again and it tried to troubleshoot it and it couldn't make it work like it tried like 10 times. And then I said, you know what? Let's do something else. Let's go back to version 3.3. And then add one component at a time that you added to version 3.4. It's like, oh, that's a great idea. Let's do this. And then we did, and we found the one thing that made it fail very, very quickly. Now, this was after it tried to do it on its own for over an hour. It was running in the background and trying different things. And every time it gives me one to install and I install it and it fails and I don't care because I'm doing five projects in parallel and I go back to it and see what's happening. It's not really quote unquote wasting my time, but it shows that you, you still have your own. Knowledge and experience and understanding of how to do things correctly in order to solve things, especially on the testing side, like building it, it knows how to build NA 10 processes better than most people. Uh, probably not better than Jeremy, a hundred percent better than me. Uh, I'm not an expert on N eight 10. So it, it builds the process, but now testing the process, you know, what great looks like, what the output needs to be, and it will give you an output. But the output may not be what you needed, or it's not optimized, or it's not optimized in these specific scenarios, and this is up to you to convey that feedback in order to then, uh, get it to be what it needs to be. So what Jeremy said to get to a, okay, now I can use this at scale in my business effectively in a way that actually generates business value. A few days, and as I said, and I'm sure it's the same for generation, you can verify that's not the only one doing these days. Right. You just go back and forth.

Jeremy G

Yeah, exactly.

Isar Meitis

Uh, any thoughts you will add that you think people can learn from as far as the, how do you approach this? How do you build these things? How do you do this for clients?

Jeremy G

Yeah. Um, if you're a beginner, if you want to do something like this, uh, as you said, you don't have to be a master.

2 GMT20260402-164029_Recording_gvo_1280x720

you need is, uh, is to have a clear idea of what you want to build. Even if that will evolve with time, with the back and forth, with the iteration, you will do in the, in the process of creating it. But if you don't know exactly what you're building, you will, you will be, you will be failing and, and, and have. You want to be the pilot in this, in this, in this ship. You don't want to be on the, on the passenger, uh, side. That's, that's the point. Uh, but then the execution of what you have in mind, you delegate to the agent. That's, that's probably better than us to do that. And it's definitely. Too annoying to do it yourself when you can automate it. So yeah, delegate these parts. Um, and yeah, yeah, that's, uh, that would be it. I would, uh, I would, uh, strongly recommend to study a little how agent works with those files so you can, you know, make sure that you have, uh, the best start possible with the, you know, not wasting time. Invest this time in the, in the first agent you do, uh, and then you will, will save your time in the next agents. You will, you will create to give it the right tools, the right skills. And then just, you know, ship something, trade something, it will be terrible. And that's okay. And, uh, and the more you you use it, the better. The better you will, uh, become. Your skills will improve. How to manage that. And your agent will become better as well. It'll improve. Save the, the practices, learn how you think and work and, uh, and you'll be saving so much time and, and, and resources. I will add one thing that connects to what you said, kind of like the, the underlying concepts behind it. All of these are very, very solid advice. The one thing that I will add, which, which connects exactly to what you said, is think infrastructure. Like every time you build something, and it won't happen the first one you build or the second or the third, but once you build five or more, you're like, oh, this thing, the way I created the MD file, uh, I can use for anything. So now you have a new infrastructure concept that you can use in every new agent that you build. The way I create NA 10 processes is now the same. It's now structured. It has the same process, the same steps, the same. So if you think about the things that you will use across multiple agents and multiple process that you build, and when you review them saying, this can be reused. Don't build it as a component in this solution, build it as an infrastructure tool that then can be used across everything. So when Jeremy's saying, oh, how does MD files needs to be set up for building agents? Once you figure it out and there's, there's probably 50 different ways to do it. But once you figure it out in a way that works for you, say, build an agent that says, every time I'm building a new agent, I want to create this file structure. It needs to have these files and these files do these things, and I want you to have these instructions as a default in them and you will do it. That's it. You don't have to worry about it anymore. So now your file structure, every single time you start something will be identical to the way it was three weeks ago because it now does. So if you think about it this way, and I'm building not. The car, I'm building the Legos, so then I can build a car and I can build a plane and I can build whatever. And after I build the Legos, now I'll build the car. Then you'll have a lot more reusable legos that you can use across, uh, different things. Yeah, great image. That's exactly it. Yeah. Awesome. Uh, Jeremy, this was fantastic. I think you, I think for a lot of people, the whole concept of agents sounds more like magic and, and, you know, alchemy than, than what it really is. And, and showing exactly this, here's, it's, it's, it's a chat in Claude in simple English, explaining what I wanna build and then it knows how to do the rest. Uh, really demystifies it and I'm sure will help a lot of people. If people want to work with, with you, follow you, hire you. What are the best ways to do that? LinkedIn is probably the best way to find me, Jeremy Ion on LinkedIn and TC nine ai, our website. Awesome. Thank you so much. This was really, really great. Thanks everybody, uh, for, for joining us and uh, thanks for having come join us every Thursday at noon.

Head Shot

Isar Meitis

Host